- New ml_signal_predictor.py with XGBoost model
- Feature extraction from historical trades (17 features)
- 5-fold cross-validation for robust training
- Integrated into enhanced_trading_check_wrapper as optional layer
- Disabled by default until model improves (AUC: 0.508)
- Key insight: Asian session is strongest predictor of success
Features used:
- Signal: confidence, threshold, regime_strength
- Session: asian/london/ny/overlap (one-hot)
- Time: hour (cyclical), day_of_week
- Quality: signal_quality score
- Direction: long/short
Usage:
- train_ml_model() to train
- enable_ml_predictor() to activate
- get_ml_status() for info
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
